Parameterized Quantum Circuits (PQC)

Overview

A parameterized quantum circuit (PQC) is the quantum “model” inside a variational algorithm. It gives you a controllable family of states or operations, indexed by parameters $\theta$.

PQCs matter in the NISQ era because:

• they can be kept relatively shallow
• they are flexible enough to represent many candidate solutions
• they connect naturally to classical optimization loops

The choice of PQC (often called the ansatz) is one of the most important design decisions in variational algorithms.

Intuition

Parameters are knobs

Think of a PQC like an experimental setup with tunable knobs:

• rotation angles
• phase shifts
• strengths of entangling interactions

Changing the knobs changes the output state.

Trainable gates intuition

A typical pattern is:

• fixed circuit template
• some gates have angles you can change

So the circuit explores a space of possible states.

Expressivity vs trainability

There is a tension:

• A very expressive ansatz can represent many states (good for finding a solution).
• But highly expressive ansätze can be hard to train (optimization can stall).

Intuition:

• If the landscape is too “flat” or too noisy, the optimizer has trouble finding a direction.
• If the ansatz is too simple, it may never represent a good solution.

Why ansatz choice matters

An ansatz is a bias. It encodes your guess about what kinds of states are useful. A well-chosen ansatz can:

• reduce depth
• reduce parameter count
• make training more stable

A poorly chosen one can:

• require too much depth (too noisy)
• be untrainable
• get stuck far from good solutions

Formal Description

We describe PQCs as parameterized unitaries.

PQC as a unitary family

A PQC can be written conceptually as:

• $U(\theta)$: a unitary that depends on parameters $\theta$

Starting from an initial state (often $|0\cdots 0\rangle$), it prepares:

• $|\psi(\theta)\rangle = U(\theta)|0\cdots 0\rangle$

You then measure some observable(s) to define a cost.

Parameter layers

Many PQCs are built from repeated layers:

• single-qubit parameterized rotations
• entangling gates
• repeat

Increasing layers usually increases expressivity but also:

• increases depth
• increases noise exposure
• increases training difficulty

Ansatz design goals (conceptual)

A good ansatz often tries to balance:

• representational power (can it reach good states?)
• hardware feasibility (can it run with low error?)
• trainability (can an optimizer reliably improve it?)

Worked Example

Consider a 2-qubit PQC template:

1. start with $|00\rangle$
2. apply a rotation on qubit 1 with angle $\theta_1$
3. apply a rotation on qubit 2 with angle $\theta_2$
4. apply one entangling gate

Even without writing gate matrices, you can see:

• $\theta_1,\theta_2$ are knobs that change the state
• the entangling gate allows the circuit to represent correlated (entangled) states

If your task requires correlation between qubits, an ansatz with no entangling gates may fail no matter how you tune angles.

Turtle Tip

Common Pitfalls

Quick Check

What’s Next

Next we define what the optimizer is trying to improve: cost functions, often built from expectation values. We’ll also discuss why measurement cost (shots) and sampling noise are central constraints in variational workflows.